Engineer Joshua Saxe: Astra useful but far from AGI hype — a 'slot machine' AI with no memory of intent

joshua_saxe · x · 2026-09-08

AI engineer Joshua Saxe reviewed Google's Astra: genuinely useful, but disappointing relative to AGI hype.

His core argument: today's models improve incrementally under a 'slot machine' paradigm — each output has what feels like an i.i.d. probability of matching your intent, with no real learning of your preferences. A substantive practitioner critique of the missing personalization and persistent memory in current models.

Original post →

More from AGI Musings

AGI Musings channel →